GPT4: In-depth Guide in 2023
Artificial Intelligence (AI) is the future trend in the tech world. AI systems are experiencing a leap forward every year, with the efforts and investments of big tech companies. ChatGPT, founded on GPT-3.5, was one of the most popular tech developments of 2022, followed by new versions.
We now have GPT4, the latest and most advanced language model at hand. Those who test with the model admit that its quality is outstanding. For you to have a better understanding of this new language model, we provide an in-depth guide focusing on its use, training, features and limitations.
What is GPT-4?
Generative pre-trained transformer 4 (GPT4) is OpenAI‘s latest language model under GPT series, released on March 14, 2023. Microsoft has confirmed that certain versions of Bing that utilize GPT technology were utilizing GPT-4 prior to its official release.1
GPT-4 is a multimodal large language model of significant size that can handle inputs of both images and text and provide outputs of text. Although it may not perform as well as humans in many real-world situations, the new model has demonstrated performance levels on several professional and academic benchmarks that are comparable to those of humans.
What is the current availability of GPT-4?
Users who subscribe to ChatGPT Plus will have access to GPT-4 on chat.openai.com, but there will be a limit on usage. According to OpenAI, the exact usage limit will be adjusted based on the level of demand and the performance of the system, but it is expected to be highly restricted due to capacity limitations.
OpenAI may consider introducing a new subscription level that allows for higher-volume usage of GPT-4, based on the observed traffic patterns. Additionally, they are planning to provide some free GPT-4 queries to allow individuals without a subscription to test the model at some point in the future.
How was GPT-4 trained?
The deep learning training of GPT-4 took place on the AI supercomputers of Microsoft Azure. Azure’s infrastructure, which is optimized for AI, also enables the distribution of GPT-4 to users worldwide.2
Similar to earlier GPT models, the GPT-4 base model was trained to anticipate the next word in a given text and was trained on a mixture of publicly available data, such as internet data, and proprietary data that we have licensed. However, GPT4 has a slightly but importantly different advantage in training:
Training with human feedback
Although the base model can generate a broad range of answers when prompted with a question, many of which may not align with the user’s intended meaning, they refine the model’s behavior through reinforcement learning with human feedback (RLHF) to ensure that it stays within certain boundaries that are consistent with the user’s objectives.
OpenAI utilized feedback from human sources, including human feedback provided by users of ChatGPT, to enhance the performance of GPT-4. They also collaborated with more than 50 specialists to obtain initial feedback in various areas, such as AI safety and security.
What were the previous models before GPT-4?
Legacy GPT-3.5
Legacy GPT-3.5 was the first ChatGPT model released in November 2022. This free version is still available to all users as of March 2023. This is the version with the lowest capabilities in terms of reasoning, speed and conciseness, compared to the following models (Figure 1).
Figure 1. The features of Legacy GPT-3.5
Source: OpenAI
Default GPT-3.5
Default GPT-3.5 is a pro version of the legacy model, released in the beginning of March 2023. This was the paid previous version before GPT4. According to OpenAI, this version has a better conciseness and is faster than the legacy version. However, there is no qualitative difference between the reasoning capabilities of the two versions (Figure 2).
Figure 2. The features of Default GPT-3.5
Source: OpenAI
GPT-4
Although it is disadvantageous in terms of its response speed, GPT-4 outperforms the earlier two versions in terms of reasoning and conciseness (Figure 3).
Figure 3. The features of GPT-4
Source: OpenAI
What are the distinctive features of GPT-4?
1- Visual input option
Although it cannot generate images as outputs, it can understand and analyze image inputs. GPT-4 has the capability to accept both text and image inputs, allowing users to specify any task involving language or vision. It can generate various types of text outputs, such as natural language and code, when presented with inputs that include a mix of text and images (Figure 4).
Figure 4. GPT-4 understanding the visual input and producing text output
Source: OpenAI
GPT-4 demonstrates similar abilities when processing input that includes both text and visual elements in various domains, including documents containing text, photographs, diagrams, or screenshots.
However, GPT4’s visual input option is not currently available to users on ChatGPT. OpenAI is working on implementing this to the chatbot.
2- Higher word limit
Figure 5. The comparison of ChatGPT with GPT-3.5 and GPT-4 in terms of word limit
Source: OpenAI
GPT-4 has the ability to process more than 25,000 words of text (see Figure 5 above), making it suitable for a variety of use cases, such as:
- Creating long-form content
- Carrying out extended conversations
- Conducting document analysis and search tasks (Figure 6)
Figure 6. GPT-4 conducts an analysis on the given link and provides a precise answer to a related question about the content
Source: OpenAI
3- Advanced reasoning capability
GPT-4 is outstanding compared to the earlier versions with its natural language understanding (NLU) capabilities and problem solving abilities. The difference may not be observable with a superficial trial, but the test and benchmark results show that it is superior to others in terms of more complex tasks.
As an example, OpenAI tested the large language models in a simulated bar exam. GPT-4’s bar exam results show that it scored in the top 10% of test-takers, while GPT-3.5’s score was in the bottom 10%.3 Overall, the performance of GPT-4 on various professional exams outperformed that of GPT-3.5 (Figure 7).
Figure 7. The comparative analysis of exam results of the three GPT models
Source: OpenAI
4- Advanced creativity
As a result of its higher language capabilities, GPT-4 is advanced in creativity compared to earlier models (Figure 7). This can make the language model more adaptive to certain use cases that require creative writing skills, such as:
- Screenplay writing
- Blog post creation
- Essay writing
Figure 8. GPT-4 produces an an output for a highly complex task that requires not only expertise but also creativity
Source: OpenAI
5- Adjustment for inappropriate requests
ChatGPT was criticized for its handicap in terms of providing answers to inappropriate requests such as explaining how to make bombs at home, etc. OpenAI was working on this problem, and made some adjustments to prevent the language models from producing such content.
According to OpenAI, GPT-4 is 82% less likely to respond to requests for disallowed and sensitive content (Figure 9).4
Figure 9. A comparison of the language models in terms of their tendency to produce responses to inappropriate requests
Source: OpenAI
6- Increase in fact-based responses
Another limitation of the earlier GPT models was that their responses were not factually correct for a substantive number of cases. OpenAI announces that GPT-4 is 40% more likely to produce factual responses than GPT-3.5.
Figure 10. A comparison of GPT models in terms of their performance to produce factually-correct responses
Source: OpenAI
7- Steerability
“Steerability” is a concept in AI that refers to its ability to modify its behavior as required. This capacity can be valuable, such as when the model needs to act as a compassionate listener, but it can also be risky if individuals convince the model that it has negative qualities, such as being malicious or depressed.
GPT-4 incorporates steerability more seamlessly than GPT-3.5, allowing users to modify the default ChatGPT personality (including its verbosity, tone, and style) to better align with their specific requirements (Figure 11).
Figure 11. GPT-4 appropriates a Socratic style during the conversation as commanded at the beginning
Source: OpenAI
What are the limitations of GPT-4?
ChatGPT usage cap limitation
GPT4, which is publicly available via ChatGPT, currently has a usage cap limitation (Figure 12). Users facing this limitation are directed to use the earlier GPT-3.5 versions.
Figure 12. GPT-4 gives the error of usage cap
Common LLM reasoning limitation
Although GPT-4 has impressive abilities, it shares some of the limitations of earlier GPT models. The model is not completely dependable, and it has a tendency to generate false information and make mistakes in its reasoning. Consequently, users should exercise caution when relying on the language model’s outputs, particularly in high-stakes situations. Depending on the specific use-case, it may be necessary to adopt various measures, such as additional human review, contextual grounding, or even avoiding high-stakes applications altogether, to ensure that the outputs are reliable.
Knowledge update limitation
Like previous GPT models, GPT-4 generally does not possess knowledge of events that have occurred after the vast majority of its training data was collected (i.e., before September 2021). Also, it does not have the ability to learn from its experiences.
Scientific research limitation
Besides ChatGPT Plus users, GPT4 is currently available to the use of software developers as an API to develop applications and systems.
However, OpenAI has disclosed less technical information about GPT-4 compared to its previous models, which has been criticized by some AI researchers.5 They argue that the lack of detailed information hinders open research into GPT-4’s potential biases and safety concerns. According to a research scientist, this is a dead-end for the scientific community.6
If you have questions or need help in finding vendors, we can help:
- “Confirmed: the new Bing runs on OpenAI’s GPT-4.” Bing Blogs, 14 March 2023, https://blogs.bing.com/search/march_2023/Confirmed-the-new-Bing-runs-on-OpenAI%E2%80%99s-GPT-4/. Accessed 27 March 2023.
- “Introducing GPT-4 in Azure OpenAI Service | Azure Blog and Updates.” Microsoft Azure, 21 March 2023, https://azure.microsoft.com/en-us/blog/introducing-gpt4-in-azure-openai-service/. Accessed 27 March 2023.
- “GPT-4.” OpenAI, 14 March 2023, https://openai.com/research/gpt-4. Accessed 27 March 2023.
- Supra note 3.
- “GPT-4 is bigger and better than ChatGPT—but OpenAI won’t say why.” MIT Technology Review, 14 March 2023, https://www.technologyreview.com/2023/03/14/1069823/gpt-4-is-bigger-and-better-chatgpt-openai/. Accessed 27 March 2023.
- “GPT-4 is here: what scientists think.” Nature, 30 December 2017, https://www.nature.com/articles/d41586-023-00816-5. Accessed 27 March 2023.
Cem has been the principal analyst at AIMultiple since 2017. AIMultiple informs hundreds of thousands of businesses (as per similarWeb) including 55% of Fortune 500 every month.
Cem’s work has been cited by leading global publications including Business Insider, Forbes, Washington Post, global firms like Deloitte, HPE and NGOs like World Economic Forum and supranational organizations like European Commission. You can see more reputable companies and resources that referenced AIMultiple.
Throughout his career, Cem served as a tech consultant, tech buyer and tech entrepreneur. He advised enterprises on their technology decisions at McKinsey & Company and Altman Solon for more than a decade. He also published a McKinsey report on digitalization.
He led technology strategy and procurement of a telco while reporting to the CEO. He has also led commercial growth of deep tech company Hypatos that reached a 7 digit annual recurring revenue and a 9 digit valuation from 0 within 2 years. Cem’s work in Hypatos was covered by leading technology publications like TechCrunch like Business Insider.
Cem regularly speaks at international technology conferences. He graduated from Bogazici University as a computer engineer and holds an MBA from Columbia Business School.
Artificial Intelligence (AI) is the future trend in the tech world. AI systems are experiencing a leap forward every year, with the efforts and investments of big tech companies. ChatGPT, founded on GPT-3.5, was one of the most popular tech developments of 2022, followed by new versions.
We now have GPT4, the latest and most advanced language model at hand. Those who test with the model admit that its quality is outstanding. For you to have a better understanding of this new language model, we provide an in-depth guide focusing on its use, training, features and limitations.
What is GPT-4?
Generative pre-trained transformer 4 (GPT4) is OpenAI‘s latest language model under GPT series, released on March 14, 2023. Microsoft has confirmed that certain versions of Bing that utilize GPT technology were utilizing GPT-4 prior to its official release.1
GPT-4 is a multimodal large language model of significant size that can handle inputs of both images and text and provide outputs of text. Although it may not perform as well as humans in many real-world situations, the new model has demonstrated performance levels on several professional and academic benchmarks that are comparable to those of humans.
What is the current availability of GPT-4?
Users who subscribe to ChatGPT Plus will have access to GPT-4 on chat.openai.com, but there will be a limit on usage. According to OpenAI, the exact usage limit will be adjusted based on the level of demand and the performance of the system, but it is expected to be highly restricted due to capacity limitations.
OpenAI may consider introducing a new subscription level that allows for higher-volume usage of GPT-4, based on the observed traffic patterns. Additionally, they are planning to provide some free GPT-4 queries to allow individuals without a subscription to test the model at some point in the future.
How was GPT-4 trained?
The deep learning training of GPT-4 took place on the AI supercomputers of Microsoft Azure. Azure’s infrastructure, which is optimized for AI, also enables the distribution of GPT-4 to users worldwide.2
Similar to earlier GPT models, the GPT-4 base model was trained to anticipate the next word in a given text and was trained on a mixture of publicly available data, such as internet data, and proprietary data that we have licensed. However, GPT4 has a slightly but importantly different advantage in training:
Training with human feedback
Although the base model can generate a broad range of answers when prompted with a question, many of which may not align with the user’s intended meaning, they refine the model’s behavior through reinforcement learning with human feedback (RLHF) to ensure that it stays within certain boundaries that are consistent with the user’s objectives.
OpenAI utilized feedback from human sources, including human feedback provided by users of ChatGPT, to enhance the performance of GPT-4. They also collaborated with more than 50 specialists to obtain initial feedback in various areas, such as AI safety and security.
What were the previous models before GPT-4?
Legacy GPT-3.5
Legacy GPT-3.5 was the first ChatGPT model released in November 2022. This free version is still available to all users as of March 2023. This is the version with the lowest capabilities in terms of reasoning, speed and conciseness, compared to the following models (Figure 1).
Figure 1. The features of Legacy GPT-3.5
Source: OpenAI
Default GPT-3.5
Default GPT-3.5 is a pro version of the legacy model, released in the beginning of March 2023. This was the paid previous version before GPT4. According to OpenAI, this version has a better conciseness and is faster than the legacy version. However, there is no qualitative difference between the reasoning capabilities of the two versions (Figure 2).
Figure 2. The features of Default GPT-3.5
Source: OpenAI
GPT-4
Although it is disadvantageous in terms of its response speed, GPT-4 outperforms the earlier two versions in terms of reasoning and conciseness (Figure 3).
Figure 3. The features of GPT-4
Source: OpenAI
What are the distinctive features of GPT-4?
1- Visual input option
Although it cannot generate images as outputs, it can understand and analyze image inputs. GPT-4 has the capability to accept both text and image inputs, allowing users to specify any task involving language or vision. It can generate various types of text outputs, such as natural language and code, when presented with inputs that include a mix of text and images (Figure 4).
Figure 4. GPT-4 understanding the visual input and producing text output
Source: OpenAI
GPT-4 demonstrates similar abilities when processing input that includes both text and visual elements in various domains, including documents containing text, photographs, diagrams, or screenshots.
However, GPT4’s visual input option is not currently available to users on ChatGPT. OpenAI is working on implementing this to the chatbot.
2- Higher word limit
Figure 5. The comparison of ChatGPT with GPT-3.5 and GPT-4 in terms of word limit
Source: OpenAI
GPT-4 has the ability to process more than 25,000 words of text (see Figure 5 above), making it suitable for a variety of use cases, such as:
- Creating long-form content
- Carrying out extended conversations
- Conducting document analysis and search tasks (Figure 6)
Figure 6. GPT-4 conducts an analysis on the given link and provides a precise answer to a related question about the content
Source: OpenAI
3- Advanced reasoning capability
GPT-4 is outstanding compared to the earlier versions with its natural language understanding (NLU) capabilities and problem solving abilities. The difference may not be observable with a superficial trial, but the test and benchmark results show that it is superior to others in terms of more complex tasks.
As an example, OpenAI tested the large language models in a simulated bar exam. GPT-4’s bar exam results show that it scored in the top 10% of test-takers, while GPT-3.5’s score was in the bottom 10%.3 Overall, the performance of GPT-4 on various professional exams outperformed that of GPT-3.5 (Figure 7).
Figure 7. The comparative analysis of exam results of the three GPT models
Source: OpenAI
4- Advanced creativity
As a result of its higher language capabilities, GPT-4 is advanced in creativity compared to earlier models (Figure 7). This can make the language model more adaptive to certain use cases that require creative writing skills, such as:
- Screenplay writing
- Blog post creation
- Essay writing
Figure 8. GPT-4 produces an an output for a highly complex task that requires not only expertise but also creativity
Source: OpenAI
5- Adjustment for inappropriate requests
ChatGPT was criticized for its handicap in terms of providing answers to inappropriate requests such as explaining how to make bombs at home, etc. OpenAI was working on this problem, and made some adjustments to prevent the language models from producing such content.
According to OpenAI, GPT-4 is 82% less likely to respond to requests for disallowed and sensitive content (Figure 9).4
Figure 9. A comparison of the language models in terms of their tendency to produce responses to inappropriate requests
Source: OpenAI
6- Increase in fact-based responses
Another limitation of the earlier GPT models was that their responses were not factually correct for a substantive number of cases. OpenAI announces that GPT-4 is 40% more likely to produce factual responses than GPT-3.5.
Figure 10. A comparison of GPT models in terms of their performance to produce factually-correct responses
Source: OpenAI
7- Steerability
“Steerability” is a concept in AI that refers to its ability to modify its behavior as required. This capacity can be valuable, such as when the model needs to act as a compassionate listener, but it can also be risky if individuals convince the model that it has negative qualities, such as being malicious or depressed.
GPT-4 incorporates steerability more seamlessly than GPT-3.5, allowing users to modify the default ChatGPT personality (including its verbosity, tone, and style) to better align with their specific requirements (Figure 11).
Figure 11. GPT-4 appropriates a Socratic style during the conversation as commanded at the beginning
Source: OpenAI
What are the limitations of GPT-4?
ChatGPT usage cap limitation
GPT4, which is publicly available via ChatGPT, currently has a usage cap limitation (Figure 12). Users facing this limitation are directed to use the earlier GPT-3.5 versions.
Figure 12. GPT-4 gives the error of usage cap
Common LLM reasoning limitation
Although GPT-4 has impressive abilities, it shares some of the limitations of earlier GPT models. The model is not completely dependable, and it has a tendency to generate false information and make mistakes in its reasoning. Consequently, users should exercise caution when relying on the language model’s outputs, particularly in high-stakes situations. Depending on the specific use-case, it may be necessary to adopt various measures, such as additional human review, contextual grounding, or even avoiding high-stakes applications altogether, to ensure that the outputs are reliable.
Knowledge update limitation
Like previous GPT models, GPT-4 generally does not possess knowledge of events that have occurred after the vast majority of its training data was collected (i.e., before September 2021). Also, it does not have the ability to learn from its experiences.
Scientific research limitation
Besides ChatGPT Plus users, GPT4 is currently available to the use of software developers as an API to develop applications and systems.
However, OpenAI has disclosed less technical information about GPT-4 compared to its previous models, which has been criticized by some AI researchers.5 They argue that the lack of detailed information hinders open research into GPT-4’s potential biases and safety concerns. According to a research scientist, this is a dead-end for the scientific community.6
If you have questions or need help in finding vendors, we can help:
- “Confirmed: the new Bing runs on OpenAI’s GPT-4.” Bing Blogs, 14 March 2023, https://blogs.bing.com/search/march_2023/Confirmed-the-new-Bing-runs-on-OpenAI%E2%80%99s-GPT-4/. Accessed 27 March 2023.
- “Introducing GPT-4 in Azure OpenAI Service | Azure Blog and Updates.” Microsoft Azure, 21 March 2023, https://azure.microsoft.com/en-us/blog/introducing-gpt4-in-azure-openai-service/. Accessed 27 March 2023.
- “GPT-4.” OpenAI, 14 March 2023, https://openai.com/research/gpt-4. Accessed 27 March 2023.
- Supra note 3.
- “GPT-4 is bigger and better than ChatGPT—but OpenAI won’t say why.” MIT Technology Review, 14 March 2023, https://www.technologyreview.com/2023/03/14/1069823/gpt-4-is-bigger-and-better-chatgpt-openai/. Accessed 27 March 2023.
- “GPT-4 is here: what scientists think.” Nature, 30 December 2017, https://www.nature.com/articles/d41586-023-00816-5. Accessed 27 March 2023.
Cem has been the principal analyst at AIMultiple since 2017. AIMultiple informs hundreds of thousands of businesses (as per similarWeb) including 55% of Fortune 500 every month.
Cem’s work has been cited by leading global publications including Business Insider, Forbes, Washington Post, global firms like Deloitte, HPE and NGOs like World Economic Forum and supranational organizations like European Commission. You can see more reputable companies and resources that referenced AIMultiple.
Throughout his career, Cem served as a tech consultant, tech buyer and tech entrepreneur. He advised enterprises on their technology decisions at McKinsey & Company and Altman Solon for more than a decade. He also published a McKinsey report on digitalization.
He led technology strategy and procurement of a telco while reporting to the CEO. He has also led commercial growth of deep tech company Hypatos that reached a 7 digit annual recurring revenue and a 9 digit valuation from 0 within 2 years. Cem’s work in Hypatos was covered by leading technology publications like TechCrunch like Business Insider.
Cem regularly speaks at international technology conferences. He graduated from Bogazici University as a computer engineer and holds an MBA from Columbia Business School.
Denial of responsibility! Techno Blender is an automatic aggregator of the all world’s media. In each content, the hyperlink to the primary source is specified. All trademarks belong to their rightful owners, all materials to their authors. If you are the owner of the content and do not want us to publish your materials, please contact us by email – [email protected]. The content will be deleted within 24 hours.